EP1382017B1

EP1382017B1 - Image composition evaluation

Info

Publication number: EP1382017B1
Application number: EP01272086.8A
Authority: EP
Inventors: Stephen Philip Cheatle
Original assignee: Hewlett Packard Development Co LP
Current assignee: Hewlett Packard Development Co LP
Priority date: 2000-12-22
Filing date: 2001-12-20
Publication date: 2018-08-22
Anticipated expiration: 2021-12-20
Also published as: JP2004526179A; WO2002052839A2; EP1382017A2; US7035477B2; US20020191860A1; WO2002052839A3

Description

The present invention relates to method of providing image composition evaluation, or advice, and to an image composition evaluation system utilising the method.
The capturing of an image, whether that image is being captured permanently on photographic film or as a digital still image, or whether the image is being captured as a moving image using a video camera or television camera or the like, nearly always first involves composing the image prior to capture. This is normally accomplished through a view finder or other similar arrangement associated with the camera apparatus. Successfully composing an image to produce a pleasing picture is a skill normally acquired over time and is a skill that inexperienced camera users often find difficult to acquire. Even experienced photographers or camera users can sometimes overlook some compositional aspects of an image due to a lack of concentration at the time of capture of the image, or due to capturing the image in a hurry.
A pleasingly composed image often follows one or more known composition rules. These rules include, for example, positioning elements of interest according to the "rule of thirds", ensuring that elements of interest are not positioned too close to the edge of the image frame, framing the main area of interest by placing areas of very little interest at the edge of the image, for example placing relatively dark areas, or areas containing relatively little activity, at the edge of the image, and positioning strong diagonal lines to run towards the main subject of the image. The "rule of thirds" involves subdividing the image area using two equally spaced horizontal lines and two equally spaced vertical lines (in the manner of a "noughts and crosses" board) and positioning elements of interest on the intersections of the horizontal and vertical lines, or equally placing strong horizontal or vertical features on one of the horizontal and vertical lines.
Although these rules of composition are well known and can be studied, inexperienced photographers often find them difficult to apply. In particular, the lack of immediate feedback concerning the proposed composition of an image makes the learning process time consuming and difficult.
It has been proposed by the present applicant in the co-pending United Kingdom patent application number GB 0031423.7 to provide a method and system for automatically cropping an image after it has been captured to improve the composition of the image. The automatic cropping method disclosed in this co-pending application is based on automatic selection of one or more of the composition rules indicated above. While this post capture analysis and cropping is very useful, a disadvantage of this method is that by cropping an image after it has been captured the resulting cropped image is either smaller than the original image or is of a lower resolution than if the subject had been correctly framed in the first place. This is because inevitably by cropping the image its size is reduced and it is therefore necessary, in order to return the cropped image to the same size as the original image, to increase the size of each picture element that comprises the image. Also, cropping an image after it has been captured does not allow the full inclusion of any objects, or areas of interest, that were only partially included in the originally , poorly composed, image. Similarly, if there is inadequate border between the areas of interest and the edge of the original image, this cannot be corrected by subsequent post capture cropping.
It would therefore be advantageous to provide an indication of how well a proposed image conforms to these well known composition rules prior to capture of that image. This can be considered as providing automatic composition advice.
The following two papers disclose retrieval and classification systems utilizing image composition as a relevant parameter:

" Image retrieval based on compositional features and interactive query specification" (Hachimura K et al; Proceedings 15th International Conference On Pattern Recognition. ICPR-2000, Proceedings Of 15th International Conference On Pattern Recognition, Barcelona, Spain, 3-7 Sept. 2000, pages 262-266 vol.4, XP010533070 2000, Los Alamitos, CA, USA, IEEE Comput. Soc, USA ISBN: 0-7695-0750-6 ). This paper discloses a sy stem in which compositional parameters are extracted from images by image segmentation and object extraction procedures. T specify a query a user first selects one of seven composition templates, each of which represents a typical composition. The system then successively generates patterns of "composition models" according to the templates selected. These composition models describe the pattern of object distribution within an image. A user iteratively selects composition models, which properly describe the composition he/she wants.
" Integrating image matching and classification for multimedia retrieval on the Web" (Hirata K et al; Multimedia Computing And Systems. 1999. IEEE International Conference On Florence. Italy 7-11 June 1999, Los Alamitos, CA, USA,IEEE Comput. Soc, US, 7 June 1999 (1999-06-07), pages 237-263, XP010342881 ISBN: 0-7695-0253-9 ) This paper discloses a technique for image classification based on color, shape and composition using the primary objects.

US 5,831,670 describes deriving three composition index numbers in respect of an image viewed by a camera, and visually suggesting changes in the direction of pointing of the camera to improve the image composition. These composition index numbers comprise:

a columnar-object (pole, tree, etc.) composition index number G1 which is a measure of verticality of such objects in the image - improvement suggestions seek to make the objects vertical;
a main-subject-position composition index number G2 which is a measure of the subject's vertical centering in the image - improvement suggestions seek to centre the subject;
a horizontal-divide (e.g. horizon) composition index number G3 which is a measure of whether a horizontal divide crosses too close to the middle of the image or slants - improvement suggestions seek to move the divide from the vertical centre and make it level.

According to a first aspect of the present invention there is provided a method of providing an evaluation of the composition of an image, the method comprising acquiring an image, processing the acquired image to evaluate its composition by applying one or more compositional rules, and providing a report concerning composition of the image such that the user has an opportunity to recompose the image prior to recording the image; wherein processing of the acquired image comprises:

analyzing the acquired image to identify regions of interest within the image; and
for each identified region of interest, applying one or more compositional rules to determine compositionally-desirable boundaries for the region;

According to a second aspect of the present invention, there is provided apparatus for evaluating the composition of an image, the apparatus comprising:

an image receiving element arranged to convert an image into an electrical image signal; and
an image processor arranged to receive said electrical image signal and process it to evaluate its composition by applying one or more compositional rules, the image processor being further arranged to provide a report concerning composition of the image such that the user has an opportunity to recompose the image prior to recording the image;

analysing the image represented by the signal to identify regions of interest within the image; and
for each identified region of interest, applying one or more composition rules to determine compositionally-desirable boundaries for the region;

The present invention will now be described, by way of example, with reference to the accompanying drawings in which:

Figure 1 schematically illustrates an image composition evaluation system according to an embodiment of the present invention incorporated within a film camera;
Figure 2 schematically illustrates an image composition evaluation system according to a further embodiment of the present invention incorporated within an electronic image capture system;
Figure 3 shows an image on which composition evaluation may be performed according to an embodiment of the present invention;
Figure 4 shows an improved composition of the image of Figure 3;
Figure 5 shows an alternative improved composition of the image of Figure 3;
Figure 6 shows the image of Figure 3 after re-sampling to reduce the number of active pixels;
Figure 7 shows the image of Figure 6 once the image has been blurred and subdivided into regions of similar appearance;
Figure 8 shows the image of Figure 7 following further region merging;
Figure 9 shows the image of Figure .8 following colour clustering;
Figure 10 shows a saliency image for the image of Figure 3;
Figure 11a, 11b and 11c illustrate potential minimum and maximum cropping boundaries; and
Figure 12 illustrates a way of determining suitable positions for crop boundaries.

Figure 1 schematically illustrates an image composition evaluation system constituting an embodiment of the present invention, in conjunction with a film camera system. A subject 1 is optically focussed by a lens system 3 which receives light from the subject and focuses it onto a view finder 5. In this particular example the view finder 5 is part of both the image composition evaluation system and the optical camera system. Using a hinged mirror 7, of the type well known from single lens reflex cameras, the image 1 is also focused on a image converter 9. The image converter 9 may, for example, be a photo-detector array. The image converter 9 converts the optical image signals into an electrical image signal. Connected to the image converter 9 is an image processor 11 that performs image composition evaluation on the received image. Should the image processor 11 determine that the received image contravenes one or more of a predefined set of composition rules, a warning signal is generated. The warning signal may be an audio warning signal produced using an audio loudspeaker 13, or alternatively the warning signal may be a visual warning signal produced, for example, by illuminating a warning lamp 15 or by giving a textural message. The visual warning may be visible through the image view finder 5. Equally, both an audio warning signal and visual warning signal may be generated. The image 1 may also be captured on photographic film 17 by the focussing elements 3 and the mirror 7. The capture of the image 1 on the photographic film 17 may take place after the image composition evaluation system has evaluated the composition of the image, thus providing the user of the camera system with the opportunity of recomposing the image 1 to finally capture a more pleasingly composed image.
Figure 2 shows an alternative arrangement of an image composition evaluation system constituting an embodiment of the present invention, in conjunction with an electronic image capture system. Unless otherwise stated, those parts that are common between the system shown in Figure 2 and the system shown in Figure 1 are numbered with identical reference numerals. An image 1 is optically focused using lens system 3 that receives the image 1 and focuses it onto an image conversion device 9 to convert the optical image into an electronic image signal. The image signal is then passed to an image processor 11 where image composition evaluation is performed. The image processor may also output the image signal to an image view finder 5 which may be implemented as a display, such as a LCD display. Alternatively, the electronic image signal from the image convertor 9 may be fed directly to the image view finder 5, as indicated by hash line 23 of Figure 2. Having evaluated the composition of the image, the image processor may provide a visual warning system in the image view finder 5 in addition to displaying the image. The visual warning system may simply be the illumination, or flashing, of a symbol, message, warning lamp or LED within the image view finder 5, or may be represented by highlighting, dimming, or otherwise indicating, those areas of the image that specifically contravene one or more of the predefined set of composition rules. The image composition evaluation may occur continuously, as just described, or may only occur in response to an image composition evaluation request signal provided from an composition evaluation request switch 19. The composition evaluation switch 19 may also serve as an image capture switch as part of an electronic image capture system that causes the image processor to capture the image 1 onto an electronic recording media 21. The image composition evaluation switch 19 may in this case be a two, or three position switch where the first position generates the image composition evaluation request signal but the switch 19 must be manipulated into a further, for example the second, position to cause image capture to occur.
In one embodiment of the present invention, the method of performing the image composition evaluation comprises identifying one or more regions of compositional significance or interest within the image and applying one or more predefined compositional rules to the identified regions. Suitable compositional rules may, for example, include the "rules of thirds", rejecting any regions of compositional significance that are too close to the edge of the image, ensuring that a "main region of interest" is always in the image, and more likely in the centre of the image, ensuring that relatively large areas containing very few regions of interest or significance are, if possible, not included in the image.
Figure 3 shows an image composed in a casual manner, perhaps by an inexperienced photographer. The girl on the swing 25 represents the main subject of the image. However, other potential regions of interest include the flowers 27 located at the bottom left hand corner of the image. Also, a serious compositional error has been made by only including part of a figure 29 shown at the right hand side of the image. The system of the present invention may, for example, issue a message indication that Figure 29 should be fully included or fully excluded from the image. Figure 4 illustrates one possible recomposition of the image that satisfies most compositional rules. In Figure 4 the main region of interest, the girl 25, is now the sole region of interest within the image, the final image having been recomposed to exclude the flowers 27 and the partial figure 29 of the original image shown in Figure 3. Figure 5 shows an alternative recomposition of the image of Figure 3 where only the partial figure 29 has been excluded. Of course other possible recompositions exist, such as recomposing the image to include the totality of the figure 29 shown in Figure 3 or recomposing to only include the flowers 27 as the sole region of interest. The final composition of the image remains the choice of the operator of the composition evaluation system. Where multiple recompositions exist, the user may instruct the image processor to present them for review. This may occur by activation of a further switch, or other input means such as existing keys or voice commands.
Various schemes are known for selecting a region of interest from an electronic image. One such scheme is described in the present applicants co-pending UK patent application number GB 0031423.7 entitled "automatic cropping of electronic images". A summary of the scheme disclosed in GB 0031423.7 will now be described for the sake of completeness with reference to the image shown in Figure 3. It will be appreciated that although only shown using a grey scale image to a reproduction, the original image of Figure 3 was in colour.
An automated image processing system has no a-priori knowledge of the subject matter of the photograph and therefore needs to process it in order to extract some form of representation which will indicate where the compositionally significant regions of the photograph lie.
The photograph 10 may have been taken with a camera having in excess of 2,000,000 active pixels. Analysing such a large number of pixels would be computationally very significant indeed. Thus prior to performing any other processing stamps, the image processor down samples the image in order to reduce the number of pixels therein. Figure 6 schematically illustrates the same image as shown in Figure 3, but after down sampling to 240 by 180 pixels. This down sampling has reduced the number of active pixels to 43,200. Following the down sampling, the down sampled image is then converted into an image having compressed colour variation whilst still retaining intensity variations. An example of such a processing is converting the image to the YCC colour space format. It should be noted that this is not the only colour space representation which could be used. Thus, the CIELAB colour space system can also be used. This system is well known, and defines a space in which the lightness L*, which is a measure of how bright a colour is, is plotted against the vertical axis, and two further measurements a* and b* are defined as linear axes with a* defining the colour from a red to green scale and the b* axis indicating colour on a blue to yellow scale. The measurements a* and b* are in the horizontal colour plane and are perpendicular to each other such that this colour system defines an orthogonal cartesian space. Each of the L*, a* and b* axis are defined in such a way that one unit on any of the scales has approximately the same "visibility" making this system both linear and isotropic as regards human perception. The L* axis has a scale from zero (black) to 100 (white) whilst the a* and b* scales range from -60 to +60 each. This system has the advantage that a colour difference of one unit has substantially the same visibility at any part of the colour space.
Following conversion of the image to a colour space, areas within the converted image having similar colour and intensity are generated and grown. This process commences by blurring the image, and then the blurred image is analysed in order to form "seed areas" that have a smooth colour and intensity. The seed areas are then grown by adding areas adjacent to the boundary of the seed areas where those adjacent areas have a sufficiently similar colour and intensity. A test is made to determine whether all of the pixels within the colour compressed image have been allocated to seed areas. If not the blur and region grow process is repeated in an iterative manner.
Figure 7 schematically illustrates the image of Figure 6 once all of the image has been blurred and assigned to regions. At this stage the image shown in Figure 7 contains approximately 2,800 regions, some 2,200 of which contain 10 or less pixels.
The image processing then continues by merging adjacent areas of the image which are separated by "weak edges". "Weak edges" are those boundaries that separate areas of the picture which have a relatively low colour or intensity differences. In other words, the regions which are close to one another within the YCC or CIELAB space. Adjacent areas with similar mean colours are then merged together, and then the image is analysed to determine if small areas, that is areas whose size is less than a threshold value, are completely enclosed by another larger area. If so, then the small area is merged into the larger area. A test may be made to determine whether the number of individual regions has fallen to below a predetermined threshold number. If it is judged that there are still too many regions, the merge can be repeated, possibly with the definition of what constitutes a weak edge being changed such that the distance in the colour space by which colours must be separated before they are regarded as sufficiently different not to be merged may be increased. Figure 8 shows the image following the region merging.
The image is further analysed in order to cluster similar colours together until such time as the number of colours has dropped to an appropriate number, which is typically in the region of 20 or so. The image of clustered colours is schematically illustrated in Figure 9.
It should be noted that as used herein a region is a spatially connected sub-area of the image. However a cluster is a collection of similar regions, but the regions do not need to be adjacent to one another.
It can be seem with reference to Figure 9 that the main part of the flowers 27 have been merged into areas of uniform colour. Similarly, the girl's face has been merged into an area of uniform colour 51 as have her trousers 52. Large areas of the background have also been merged into areas of substantially uniform colour, for example the tree towards the left hand side of the image.
Next an interest metric is formed on the basis of the unusualness of the colour, and the image is analysed to determine the compositionally significant properties therein from amongst a plurality of different possible properties.
One such analysis that may be performed is the analysis of the clustered colours shown in Figure 9 to determine how unusual they are. The image shown in Figure 9, as noted hereinbefore, comprises approximately 20 or so different colour clusters. These clusters are then sorted in order to identify how many pixels belong to each one of the colours.
Each of the colour clusters is processed in turn. When a colour is processed, the colour distance between it and each of the other colour clusters is calculated, the clusters are then sorted in order of colour distance from the colour cluster being processed. A cumulative histogram can then be formed for the colour cluster under test, by counting the cumulative sum of image pixels which are included in an increasing number of clusters along the colour distance dimension.
Clusters which, together with closely coloured neighbouring clusters, occupy a relatively large proportion of the pixels of the image are deemed to be background. Conversely, cluster colours which together with closely coloured neighbouring clusters occupy only a relatively small proportion of the pixels of the image are deemed to be foreground. By this analysis, cluster colours can be allocated a default saliency based on the likelihood that they are foreground colours.
However, colour mapping is not the only process that is applied in order to determine a saliency image. In general, those regions which are located towards the edges of the image may be penalised as they may belong to objects which are not fully in frame.
Further processes, such as pattern recognition may also be applied to the image. Thus, a search may be made to identify bodies or faces as a result of comparing areas within the image against models held within a model library.
Figure 10 schematically illustrates a saliency image of Figure 3 following the conclusion of the one or more analysis processes.
The saliency image is processed to subdivide it into a small number of large areas (typically rectangles) which enclose the majority of the saliency in the image. Thus, the selected areas enclose the bright regions of the saliency image. One method of doing this is to form the sums of saliency pixel values along each row, and separately, down each column. Plotting these sums against the vertical and horizontal axes respectively, shows the vertical and horizontal distributions of saliency. These can then be analysed to find the widest minimum in either the vertical or horizontal saliency distribution. The image can then be split into three parts at this minimum. A first part comprises a horizontal, or as the case may be vertical, band through the image having a width substantially corresponding to that of the minimum. This part can be ignored as non salient. This will then leave two parts of the image each side of this minimum band which will contain saliency (except in the case where the minimum band is adjacent one of the edges of the image in which case there will only be one non-empty or salient side). These parts can each be processed by the same algorithm. The part with the widest minimum can be split in an analogous manner, discarding the width of the minimum and hence splitting that part into two smaller parts. This process can continue with each stage splitting the part about the best minimum until one of the following limiting conditions is reached:

a. No minimum can be found in any of the remaining parts. I.e. no minimum is found which is sufficiently wide and sufficiently low in saliency.
b. The fraction of the total saliency of the image which is outside of the retained block reaches some predetermined limit, such as 5%.

The result of this process is that a small set of rectangular blocks which enclose the major areas of saliency of the image are derived.
Suppose that the image is initially framed such that, as shown in Figure 3, it includes unwanted features.
Once features relevant to the composition of the image have been identified, the saliency map can now include regions of the image which are defined as include regions and exclude regions. Thus, considering Figure 11A the girl has been identified as an "include" region and has been framed by a crop boundary 60 which represents the minimum boundary possible to include all of the girl therein. Similarly, the flowers have been identified as an include region and have been framed by a crop boundary 61 representing the minimum crop required to include the flowers. Furthermore, "must exclude" regions have been identified and enclosed by crop boundaries 64 and 66 respectively. However, these regions may be highlighted to a user such that he is given the opportunity to recompose the image in such a way that these features are fully included.
Thus, at this point, and optionally in response to preferences set by the user, the user may be presented with an image corresponding to or based on that shown in Figure 10 where compositionally significant regions are highlighted; or an image similar to that of Figure 11a where minimum crop boundaries are represented. Indeed such images could be colour coded such that the include regions are presented in one colour, for example green, whereas the exclude regions are presented in another colour, such as red.
The exclude regions may be significant and for example, it may be desired to re-frame the image such that the partial Figure 29 is fully framed within the image. This compositional option can be given to a user of the present invention. However, it is self evident that the Figure 29 could not be included in a system which utilised only post capture image processing.
Other problems might also be indicated at this point, such as the inclusion of large boring areas, badly placed horizons, the image being tilted, people looking out of frame, the camera pointing directly into the sun, edges with unnecessarily high levels of activity, and so on. This gives the user the opportunity to recompose the photograph.
As a further alternative, the user may be presented with suggested recompositions of the image. In order to do this, some further processing is required. An example of these additional processes is given below.
Having identified the minimum crop boundary, it is then advantageous to identify the maximum crop boundary. With regards to Figure 11B, one potential maximum crop boundary 68 has been identified. This crop boundary abuts the must exclude regions 64 and 66, but also abuts the edge of the include region 61. The boundary also extends between the upper and lower edges of the photograph. This crop boundary 68 represents the maximum crop boundary available to include the girl but to exclude the flowers. However, an alternative crop boundary is available which includes both the girl and the flowers. Thus, as shown in Figure 11C a further minimum crop boundary 70 can be defined which includes both the girl and the flowers (with partial exclusion of the flowers being allowed because they are so close to the edge), and a further maximum crop boundary 72 has also been defined which extends to the upper and lower edges of the photograph, to the left hand edge, but abuts the must exclude regions 64 and 66 at the right hand edge thereof.
The saliency map is analysed in order to determine how many areas of interest exist therein. Thus, if the saliency map shows N distinct areas of interest (for example areas of interest separated by some area of non-interest as determined by some adaptively set threshold) possible minimum cropping rectangles can be generated which contain alternative combinations of between 1 and N areas of interest where the minimum cropping rectangle contains a selected combination of areas of interest and excludes other areas. Thus this corresponds to generation of minimum cropping rectangle 60, 61 and 70 in Figures 11A and 11C. It should be noted that not all combinations may be possible as they may not be contained within a single rectangle that excludes one or more of the non-selected areas. The maximum cropping rectangle for the each single or combination of areas of interest is the maximum rectangle which contains the areas of interest but excludes the non-selected areas of interest. Thus this corresponds to rectangles 68 and 72 in Figures 11B and 11C.
Each minimum cropping rectangle 60, 61 and 70 and its associated maximum cropping limit (of which only cropping limits 68 and 72 are shown in Figures 11B and 11C) is then processed in turn. However, some initial sorting may reduce the processing required. One of the compositional rules may require that a large well centred interesting area in the image is deemed to be essential. If we apply this rule then only minimum cropping boundaries 60 and 70 are permitted, with the flowers as defined by crop boundary 61 being excluded. The first step is to select a first one of the minimum cropping boundaries 60 and 70 as a potential cropping candidate, together with its cropping limits. As search is then made to identify possible edge locations for each of the edges.
Supposing that minimum and maximum crop rectangles have been defined, and that it is now desired to find the position of suitable crop boundaries between the minimum and maximum limits. For the purpose of this description, we are going to locate the edge of one boundary, occurring to the left hand side of the minimum crop rectangle. Given that the digital image can be considered as consisting of a plurality of columns, the left hand edge of the maximum crop rectangle is located in column P, whereas the left hand edge of the minimum crop rectangle is located in column Q. Columns P and Q are not adjacent.
Sequentially each of the columns between P and Q is examined in turn in order to generate a metric of how good that column would be as a border of the cropping rectangle. Thus, the metric is constructed such that dark areas or slowly changing pixels along the column incur a low cost penalty, whereas brighter areas or alternatively rapidly changing colours in a row of pixels achieve a high penalty rating. Furthermore, the rating may also be modified with regards to the proximity of that column to the minimum and maximum crop boundaries, or indeed the proximity of that column to the edge of the picture.
In a preferred embodiment of the present invention, the edge quality metric is a function of:

i. Brightness. That is dark edges are preferred and hence incur only a low penalty.
ii. Activity. That is the sum of the colour differences between regions crossed by a row or column is analysed, with low sums scoring a lower penalty.
iii. Saliency. That is the sum of the saliency values for pixels in the row or column is formed, with low saliency incurring a lower penalty.
iv. Distance from strong colour transitions parallel-to, and on the inside of, the column or row being tested. The distance should not be too close nor too far and a weighted distance term is used to accomplish this. This latter criteria is used to avoid cropping too close to a feature, even if it is not part of the minimum cropping rectangle.

These factors are independently smoothed and normalised before being combined in order to form a weighted sum to generate the edge quality metric as shown in Figure 12.
Thus for each one of the individual columns, a penalty measurement is formed, and the penalty measurement can then be plotted with respect to column thereby obtaining a penalty measurement profile 90. The profile 90 can then be examined to determine the position of minima therein, such as broad minima 92 or the sharper minima 94 and 96 which are then deemed to be potential cropping boundaries. This process can be repeated for each of the left, right, bottom and top crop boundaries individually, and may be repeated on a iterative basis such that for example those pixels in the column which lie above the upper crop limit or below the lower crop limit are excluded from the next iteration of the crop boundary. These candidate crops can then be subject to further constraints. In practice, there will be too many constraints to satisfy all of the constraints simultaneously. Constraints may include implementing the "rule of thirds" in respect of the horizon line. Similarly. the "rule of thirds" can be introduced to act on the main feature of interest to place it ⅓ of a distance from the edge of the crop.
The final crop is also be constrained by the aspect ratio of the camera.
Once a crop candidate has been identified, it is then evaluated by applying one or more rules. Each rule is implemented as a heuristically evaluated measure on the image.
Heuristic measures are used for compositional rules such as eliminating distractions close to the edge of the frame, minimum edge quality, a preference for dark or low activity boundaries, and so on.
The combination of different rule penalties by a weighted sum allows some rules to be considered as more important than others. Again, the weightings are determined heuristically.
Other known methods of identifying regions of interest from electronic image may equally be applied to embodiments of the present invention.
Thus, as noted hereinbefore, the different compositional rules used may have different weightings associated with them to vary the importance of those rules for example, particular attention may be paid to identify large boring areas, distractions at the edges of the image, or horizon lines that are centrally placed or placed very close to the top or bottom of the image frames. The weightings may vary depending on image content.
A criterion that may be attributed particular significance may be that of identifying regions of interest that extend beyond the edge of the frame. The user of the image evaluation system may be advised by means of the audio or visual warning signals to attempt to fully capture these features, or if the region of interest identified by the image composition evaluation system is actually a combination of multiple objects, to reposition the image capture system such that these two objects are not aligned. By following this advice it is more likely to produce an image composition where the true subject is well isolated from competing regions of interest, for example by being separated by background regions.
Where candidate crops are presented to a user, and the crops lie wholly within the original badly composed image, the user can be instructed to zoom and pan in order to match the fully framed image to the suggested crop. In order to achieve this an image of the suggested crop may be faintly displayed on the viewfinder in addition to the "current" image seen by the camera.

Claims

A method of providing an evaluation of the composition of an image, the method comprising acquiring an image, processing the acquired image to evaluate its composition by applying one or more compositional rules, and providing a report concerning composition of the image such that the user has an opportunity to recompose the image prior to recording the image; characterized in that processing of the acquired image comprises:
analyzing the acquired image to identify regions of interest within the image; and

for each identified region of interest, applying one or more compositional rules to determine compositionally-desirable crop boundaries for the region;
said report being indicative of the regions of interest and their compositionally-desirable crop boundaries.
A method according to claim 1, wherein said analysis is performed in response to a composition evaluation request signal.
A method according to claim 1, wherein said report is presented visually.
A method according to claim 3, wherein said report is displayed on an image viewer.
A method according to claim 1, wherein said analyzing of the acquired image includes deriving a saliency map of the image on the basis of unusualness of one of the colour and brightness in the image.
A method according to claim 1, wherein providing the report includes presenting, as the compositionally-desirable boundaries of a region of interest, a suggested crop of the image, this crop being presented on an output device.
A method according to claim 1, wherein the candidate crop presented in the report is presented visually.
A method according to claim 1, wherein said analyzing of the acquired image identifies regions for exclusion, the determination of compositionally desirable boundaries for the regions of interest comprising for each such region:
determining a minimum crop boundary around the region of interest;

determining a maximum crop boundary around the region of interest, this maximum crop boundary excluding any region identified for exclusion but potentially including one or more other regions of interest;

determining candidate crop boundaries between the minimum and maximum crop boundaries; and

evaluating the crop candidates using said one or more compositional rules.
Apparatus for evaluating the composition of an image (1), the apparatus comprising:
an image receiving element (9) arranged to convert an image into an electrical image signal; and

an image processor (11) arranged to receive said electrical image signal and process it to evaluate its composition by applying one or more compositional rules, the image processor being further arranged to provide a report concerning composition of the image such that the user has an opportunity to recompose the image prior to recording the image;
characterized in that the image processor (11) is arranged to process the said electrical image signal by:
analysing the image represented by the signal to identify regions of interest within the image; and

for each identified region of interest, applying one or more composition rules to determine crop boundaries for the region;
said report being indicative of the regions of interest and their compositionally-desirable crop boundaries.
Apparatus according to claim 9, wherein said image processor (11) is arranged to perform its processing in response to receiving a composition evaluation request signal.
Apparatus according to claim 10, wherein said composition evaluation request signal is generated in response to the actuation of a composition evaluation request switch (19) of the apparatus.
Apparatus according to claim 11, wherein said composition evaluation request switch (19) comprises an image capture switch arranged to activate an image capture system to capture the image represented by the electrical image signal.
Apparatus according to claim 9, wherein the apparatus is arranged to provide said report visually.
Apparatus according to claim 13, wherein the apparatus is arranged to present said report on a view finder (5) of the apparatus.
Apparatus according to claim 9, wherein the image processor (11) is arranged to analyze the acquired image by deriving a saliency map of the image on the basis of unusualness of one of the colour and brightness in the image.
Apparatus according to claim 9, wherein the image processor (11) is arranged to provide the report by presenting, on an output device (5) of the apparatus, a suggested crop of the image as the compositionally-desirable boundaries of a region of interest.
Apparatus according to claim 16, wherein said output device is a visual display (5).
Apparatus according to claim 9, wherein the image processor (11), in analyzing the image, is arranged to identifies regions for exclusion; the image processor (11) being arranged to determine compositionally desirable boundaries for a said region of interest by:
determining a minimum crop boundary around the region of interest;

determining a maximum crop boundary around the region of interest, this maximum crop boundary excluding any region identified for exclusion but potentially including one or more other regions of interest;

determining candidate crop boundaries between the minimum and maximum crop boundaries; and

evaluating the crop candidates using said one or more compositional rules.
A motion picture camera including apparatus according to any one of claims 9 to 18.
A still picture camera including apparatus according to any one of claims 9 to 18.